AI Data Centre Infrastructure: GPU, HPC & Liquid Cooling

Overview

AI and HPC workloads are pushing rack densities from single-digit kilowatts towards 50-150 kW per rack. This expert course explains what that means for power distribution, liquid cooling, structural loads, networking and operations, and how facilities are adapted or purpose-built for GPU clusters.

Learning Outcomes

On completion of this course, you will be able to:

• Explain the power and thermal behaviour of GPU-based AI workloads
• Plan power distribution for 50-150 kW racks
• Compare direct-to-chip, rear-door and immersion liquid cooling
• Describe CDU, manifold and facility-water interfaces, coolant quality and leak detection
• Define rack-to-facility interfaces for power, liquid and network
• Plan commissioning, monitoring and operation of liquid-cooled halls

Module Outline

1. Module 1: AI and HPC workloads and their infrastructure demands
2. Module 2: GPU servers and rack-scale systems
3. Module 3: High-density power distribution and busway
4. Module 4: Air cooling limits and liquid cooling options
5. Module 5: Direct-to-chip cooling, CDUs and facility water loops
6. Module 6: Immersion and rear-door heat exchangers
7. Module 7: Coolant quality, leak detection and serviceability
8. Module 8: Structural, weight and space considerations
9. Module 9: Commissioning and testing of liquid-cooled systems
10. Module 10: Operating AI halls: monitoring, telemetry and incident response

The module outline is indicative and may be refined as materials are updated.